Machine learning techniques for structural health ?· Machine learning techniques for structural health…
Post on 28-Jun-2018
Machine learning techniques for structural health monitoring
Kay SMARSLY, Kosmas DRAGOS and Jens WIGGENBROCK
Chair of Computing in Civil Engineering, Bauhaus University Weimar, Coudraystr. 7, 99423 Weimar (Germany) email@example.com
Key words: Structural health monitoring, machine learning, sensor fault detection, analytical
redundancy, computer-aided structural assessment
Data-driven approaches are particularly useful for computer-supported assessment of civil
engineering structures (i) if large quantities of sensor data are available, (ii) if the physical
characteristics of the structure are complex to model (or even unknown), or (iii) if the
computational efforts are to be reduced. This paper, upon a classificational review of
machine learning techniques in structural health monitoring, reports on an embedded
machine learning approach for decentralized, autonomous sensor fault detection in wireless
sensor networks, facilitating reliable and accurate structural health monitoring. Based on
decentralized artificial neural networks, the embedded machine learning approach is applied
to perform autonomous detection of sensor faults injected in the acceleration response data
collected by a prototype structural health monitoring system. As demonstrated through
laboratory tests, the results highlight the ability of the embedded machine learning approach
to autonomously detect sensor faults in a decentralized manner, thus enhancing the
reliability and accuracy of structural health monitoring systems.
Advancements in sensor technologies have enabled economically affordable sensor
installations for long-term monitoring of civil engineering structures. Structural health
monitoring involves installations of hundreds to thousands of sensors to collect valuable data
about the structure. With increasing complexity and heterogeneity of sensor data, data
integration and data analysis have become important issues for decision making with respect
to diagnosis of the structural condition and the prognosis of structural damage [1, 2].
Data analysis in structural health monitoring, from a computer science perspective, aims at
transforming sensor data into useful information and probably into knowledge about the
structure. The information and knowledge gained from the sensor data is then used for
structural assessment and for decision making in several respects, such as life-cycle
management  or lifetime prediction . Two general approaches exist for assessing the
structural condition of civil engineering structures, physics-based approaches and data-driven
approaches . Physics-based approaches establish first-principle models, mapping the
physical characteristics of the structure (e.g. using finite element analysis), and then compare
the outputs of the physical models with sensor data obtained from the monitored structure in
order to assess the structural condition . Although significant efforts have been undertaken
to render physics-based models more efficient in terms of computational performance, for
example for embedment into resource-constraint wireless sensor nodes [7, 8], physics-based
8th European Workshop On Structural Health Monitoring (EWSHM 2016), 5-8 July 2016, Spain, Bilbao
approaches are generally more computationally intensive than data-driven approaches.
Data-driven approaches also establish models for comparison with sensor data, but data-
driven models exploit information from previously collected sensor data, referred to as
training data . While physics-based approaches are valid in a large operating range
without the need for extensive quantities of sensor data, data-driven approaches allow
learning patterns in the sensor data without any knowledge on the physical characteristics of
the structure . Data-driven approaches are particularly useful, if (i) large quantities of
sensor data are available, (ii) the physical characteristics of the structure are complex to
model (or even unknown), or (iii) the computational efforts are to be reduced.
A variety of data-driven approaches, particularly machine learning techniques, has been
proposed in structural health monitoring (SHM) for assessing civil engineering structures.
Machine learning in the context of SHM can be described as the task of generating
knowledge from past experiences (or, more precisely, from collected sensor data), focusing
on the prediction of new sensor data. While in artificial intelligence research machine
learning techniques have been studied since many decades (e.g. for robot control, human-
computer interaction, or speech recognition), its importance in SHM applications
substantially continues to grow since about 20 years [11, 12]. For example, Worden and
Manson  have illuminated the utility of machine learning to damage identification,
concluding that neural networks are still popular, and systems like support vector machines
are beginning to appear more regularly. Figueiredo et al.  have investigated auto-
associative neural networks, factor analysis, Mahalanobis distance, and singular value
decomposition to study operational and environmental variability and its influence on
damage detection of civil engineering structures. Dervilis , centered on SHM of wind
turbine blades, also explores auto-associative neural networks and formulates pattern
recognition algorithms. In addition, robust multivariate statistical methods are introduced to
account for the influence of operational and environmental variation on damage-sensitive
features; the algorithms described are the Minimum Covariance Determinant Estimator and
the Minimum Volume Enclosing Ellipsoid. Park et al. , also focusing on wind energy
research, couple Gaussian Discriminative Analysis and Gaussian Mixture Models to analyze
and to predict wind turbine loads in various atmospheric conditions. Nick et al. ,
reporting significant trade-offs between accuracy and runtime of the machine learning
techniques proposed, have used unsupervised learning for identifying the existence and
location of damage (k-means and self-organizing maps) and supervised learning for
identifying the type and severity of damage (support vector machines, naive Bayes
classifiers, and feed-forward neural networks).
This paper presents an embedded machine learning approach for decentralized,
autonomous fault detection in wireless SHM systems. Sensor faults and miscalibrations
substantially affect sensor data and may compromise the reliability and accuracy of SHM
systems. Specifically in data-driven approaches, the integrity of the sensor data needs to be
preserved to enhance the reliability and accuracy of SHM system outputs as well as the
robustness of algorithms implemented for structural health monitoring. In the study reported
in this paper, the efficient detection of sensor faults and miscalibrations is based on the
correlations among the response data of different sensor nodes, referred to as analytical
redundancy, which is implemented through an embedded machine learning approach based
on artificial neural networks. This paper is organized as follows: First, an overview of
machine learning techniques commonly used in structural health monitoring is provided.
Then, the embedded machine learning approach for decentralized, autonomous sensor fault
detection, based on artificial neural networks, is implemented into a wireless SHM system.
Serving as a testbed for the proposed approach, a laboratory test structure is used in this paper
for validation, followed by a concise summary of the study presented herein.
2 AN EMBEDDED MACHINE LEARNING APPROACH FOR DECENTRALIZED,
AUTONOMOUS SENSOR FAULT DETECTION
In computer science and in computational engineering, the process of detecting patterns
and structures within data sets is commonly known as data mining. The detection of patterns
enables future predictions and decision making, while representing the patterns in terms of
structures facilitates the extraction of conclusions on the patterns. In data mining, the
techniques employed to detect patterns within data sets fall into the category of machine
As mentioned previously, due to the computational burden of physics-based approaches in
structural health monitoring, data-driven approaches, such as machine learning, have been
gaining increasing attention. In SHM, machine learning is understood as the task of
generating knowledge about the structural behavior from previously collected sensor data.
While structural responses are theoretically well explained and documented, the detection of
such responses in full-scale structures is non-trivial due to the complex nature of actions and
the actually unknown properties of the structure. Furthermore, SHM outputs may be affected
by sensor faults and miscalibrations, which may be hardly visible in the collected data. In this
context, machine learning is applied to detect such hidden, non-evident, or inadequately
described phenomena. In this section, the machine learning techniques typically applied in
SHM are briefly discussed. Then, an embedded machine learning approach for decentralized,
autonomous detection of sensor faults and miscalibrations is presented.
2.1 Classification of machine learning techniques for structural health monitoring
Machine learning techniques can be classified into three broad categories according to the
nature of learning: 1) supervised learning, 2) unsupervised learning, and 3) semi-supervised
learning . Supervised learning provides a learning scheme with labeled data, i.e.
examples that include specified outputs (pairs of input data and output data). Using labeled
data, rules are developed in an attempt to classify new data sets. Unsupervised learning
encompasses the detection of patterns within the data sets consisting of unlabeled data, i.e.
data sets with unspecified outputs, which fit to a general rule and can, therefore, be grouped
together. From an SHM viewpoint, unsupervised learning can be used, e.g., for detecting the
existence of damage through clustering of structural response data, while supervised learning
can advantageously be employed to detect the type and severity of damage . Semi-
supervised learning, representing a combination of the two aforementioned learning schemes,
typically aims at obtaining a classification of data using both labeled and unlabeled data.
Semi-supervised learning schemes have been applied combined with other monitoring
techniques to extract information on modal characteristics of bridges .
Since most SHM problems require inferring a function from labeled training data (e.g. to
assess the data or to predict new data), supervised learning is an appropriate means to solve
these problems. In supervised learning, the algorithms, according to , can be categorized
as logic-based algorithms (e.g., decision trees and rule-based classifiers), perceptron-based
algorithms or neural networks (e.g., single-layered perceptron, multi-layered perceptron and
radial basis function networks), statistical learning (e.g., naive Bayes classifiers and Bayesian
networks), instance-based learning (e.g., k-nearest neighbor algorithm), and support vector
2.2 Prototype implementation of the machine learning approach
In this study, decentralized autonomous sensor fault detection is based on the principle of
analytical redundancy : Instead of physically installing multiple sensors for measuring
one single parameter, analytical redundancy takes advantage of the redundant information
inherent in the SHM system and utilizes the coherences and relationships between the sensors
installed in the structure. It has been proven that the peak amplitudes of the frequency
spectrum, obtained by the Fourier transformation of acceleration response data,
corresponding to resonant response (i.e. modal peak amplitudes) from different sensors of the
same structure are correlated . This correlation can be exploited to predict the modal peak
amplitudes of selected sensors, using the modal peak amplitudes of correlated sensors as
input data. Deviations between expected amplitudes and actual amplitudes (i.e. from the
measured data) are indicative of sensor faults and miscalibrations. Importantly, no a priori
knowledge about the structure or about the sensor instrumentation is required because, as a
purely data-driven approach, previously collected sensor data is taken as the sole basis for
A wireless SHM system is designed that comprises wireless sensor nodes, each of which
including an integrated 3-axis accelerometer, a base station, and a host computer. The
monitoring tasks executed by the SHM system are illustrated in Figure 1. During operation,
acceleration response data is sampled by each sensor node and locally transformed into the
frequency domain via an embedded Cooley-Tukey FFT algorithm. A peak detection
algorithm selects the highest peak of the frequency spectrum corresponding to the
fundamental eigenfrequency (modal peak amplitude), and the modal amplitudes are
communicated among the sensor nodes. Each sensor node predicts the modal amplitude of its
own acceleration response data (expected amplitude) using the modal peak amplitudes of
correlated sensor nodes and decides upon the existence of sensor faults based on deviations
between the expected and the actual modal peak amplitude. The outcomes of the fault
detection procedure of the sensor nodes are transmitted to the host computer via the base
station for storage and decision making.
Figure 1. Decentralized, autonomous fault detection procedure executed by the wireless SHM system
The decentralized autonomous fault detection procedure proposed in this study relies on
the relationships among the modal peak amplitudes from different sensors. To map these
relationships an embedded machine learning approach with a supervised learning scheme is
introduced. To this end, artificial neural networks (ANNs) are designed and distributedly
embedded into each sensor node. As shown in Figure 2, the ANNs consist of three layers of
neurons: 1) an input layer of k neurons, 2) a hidden layer of m neurons to account for the non-
linear relationship among the modal peak amplitudes of different sensors , and 3) an
output layer of one neuron, which represents the predicted modal peak amplitude of the
sensor under consideration. The data is propagated through the ANN via the synapses (i.e.
connections between neurons), based on the weight of each connection. During the ANN
training, the weights of the synapses are adjusted until a selected set of input data results in
the desired output data. The ANN properties (i.e. ANN topology and neuron behavior) are
determined based on computational steering and trial-and-error tests. For further details, the
interested reader is referred to [9, 22, 24, 25].
Figure 2. Schematic of the artificial neural network embedded into the wireless sensor nodes
3 VALIDATION OF THE MACHINE LEARNING APPROACH
Validation tests to showcase the ability of the embedded machine learning approach are
performed on a laboratory test structure. In the first part of this section, the laboratory test
setup is described. In the remainder of this section, the training of the ANN and the
determination of the ANN properties are presented. Finally, the application of the embedded
machine learning approach is illuminated.
3.1 Laboratory test setup
To validate the embedded machine learning approach, the wireless sensor nodes are
installed on the test structure, as shown in Figure 3. The test structure is a 4-story frame
structure consisting of steel plates of 250 mm x 500 mm x 0.75 mm. The plates are mounted
on threaded rods with a vertical clearance of 23 cm. At the bottom of the structure, the rods
are fixed into a solid block of 400 mm x 600 mm x 300 mm. A total of four wireless sensor
nodes, labeled A, B, C and D, are placed on the structure at the center of each story.
In addition to the wireless sensor nodes, a base station, connected to a local computer, is
placed next to the test structure.
Figure 3. Schematic of the laboratory test structure
3.2 Training and determination of the artificial neural network properties
Preliminary tests are conducted to determine the ANN properties . Several
combinations of topologies and neuron behaviors are tested. The determination of the
properties is based on the performance of the ANN in terms of time required for training and
on the output accuracy. The output accuracy (or the predictive power) is expressed through
the root mean squared error between the expected and the actual amplitudes, as shown in Eq.
1. For training, 100 sets of 4 modal peak amplitudes (from all sensor nodes) are created.
Following the standard practice in ANN training, the data set is divided to 80% training sets
to establish the relationship between inputs and outputs, 10% validation sets to decide when
to stop training, and 10% test sets to check the predictive power of the trained ANN.
iactual,iexpected,RMS FFN 1
In Eq. 1, RMS is the root mean squared error, N is the number of testing sets, Fexpected is the expected modal peak amplitude, Factual is the actual amplitude, and 1 is the fundamental eigenfrequency. The sets of modal peak amplitudes are split into three inputs and one output;
the modal peak amplitudes of sensor nodes A, C, and D are used as input to predict the modal
peak amplitude of sensor node B. Therefore, each of the tested ANNs has three neurons in
the input layer and one neuron in the output layer. Between the input layer and the output
layer, several hidden layers with varying number of neurons per hidden layer are tested. In
terms of neuron connections, both interlayer connections (between adjacent neurons) and
supralayer connections (i.e. between distant neurons) are tested. Finally, for neuron behavior,
both backpropagation and resilient backpropagation algorithms are applied. The results of the
preliminary tests are presented in Table 1.
Neuron behavior Topology Neurons per
3-1 4 6.6 0.149
3-2-1 6 13.0 0.102
3-3-1 7 17.2 0.144
3-5-1 9 25.0 0.081
3-7-1 11 32.2 0.063
3-2-2-1 8 21.0 0.092
3-5-5-1 14 46.6 0.137
3-3-1 7 15.2 0.147
3-5-1 9 22.6 0.132
3-2-2-1 8 19.4 0.137
3-3-1 7 113.0 0.153
3-5-1 9 172.4 0.143
3-2-2-1 8 120.6 0.208
Table 1. Results of preliminary tests to determine the ANN properties (source: )
The results of the preliminary tests show that all combinations of ANN properties
demonstrate satisfactory output accuracy. However, in terms of performance the time
required for training varies significantly. As a trade-off between the time and the output
accuracy an ANN with 3-2-1 topology, interlayer connections, and backpropagation neuron
behavior is selected. In the next subsection, the application of the selected ANN to detect
sensor faults injected into the acceleration response data is presented.
3.3 Application of the machine learning approach for autonomous fault detection
Two of the most common fault types, bias and precision degradation, are simulated and
injected into the acceleration response data. A bias (Figure 4a) is a deviation between the
actual response and the expected response by a constant value; precision degradation (Figure
4b) is a contamination of the response data with excessive-variance white noise. Both faults
have a noticeable impact on the modal peak amplitudes of the acceleration response data.
Figure 4. Manifestations of bias (a) and precision degradation faults (b)
Figure 5. Impact of the simulated and injected sensor faults on the modal peak amplitudes
Bias is injected by rotating one sensor node by 45o, while precision degradation is injected
by contaminating the acceleration response data of the sensor nodes with a random Gaussian
time series. Similar to the preliminary tests, the modal peak amplitudes from sensor nodes A,
C, and D, (as depicted in Figure 3) are used to predict the modal peak amplitude of sensor
node B. A threshold for the RMS at = 0.15 is established from trial-and-error tests. The results of the ANN application are summarized in Table 2.
Root mean square error No fault
RMS 0.102 0.603 0.807
Table 2. Fault detection of simulated sensor faults, indicated by root mean square error.
As shown in Table 2, the root mean squared error for both simulated sensor faults
significantly exceeds the predefined threshold. It can be concluded that fault detection using
the proposed embedded machine learning approach is a promising tool to enhance the
reliability and accuracy of monitoring.
4 SUMMARY AND CONCLUSIONS
A broad wealth of data-driven approaches, particularly machine learning approaches, has
been proposed in structural health monitoring for assessing the condition of civil engineering
structures. In machine learning approaches for structural health monitoring, the learning
scheme can be categorized into supervised, unsupervised, and semi-supervised learning.
Based on supervised learning, an embedded machine learning approach for decentralized
autonomous fault detection has been presented in this paper. The proposed approach makes
use of the analytical redundancy, i.e. the redundant information obtained by the sensors.
More specifically, the inherent correlations among the amplitudes at peaks of the frequency
spectra of acceleration response data obtained from different sensors are utilized. The modal
peak amplitude of each sensor is predicted using the modal peak amplitudes of correlated
sensors as input data. Deviations between the expected amplitude (i.e. the amplitude obtained
from the prediction) and the actual amplitude are indicative of sensor faults. To map the
relationship among the modal peak amplitudes of correlated sensor nodes, artificial neural
networks have been distributedly embedded into the wireless sensor nodes.
Validation tests have been conducted on a 4-story laboratory test structure. A total of four
wireless sensor nodes have been used, each of which placed at the center of one story.
Preliminary tests have been performed to determine the properties of the ANN, based on time
and output accuracy criteria, in which the modal peak amplitudes of the sensor nodes of three
stories have been used to predict the modal peak amplitude of the sensor node of the
remaining story. Then, two common sensor faults have been injected into the acceleration
response data of one sensor node. Finally, the ANN has been applied, and, using the modal
peak amplitudes of the other three sensor nodes the faults have been successfully detected. In
conclusion, the results of the validation tests showcase the ability of the proposed machine
learning approach to detect sensor faults. Future work could include establishing a solid
threshold to distinguish non-faulty from faulty operation as well as implementing the
automated adaptation of the fault detection procedure to account for structural changes.
Financial support of the German Research Foundation (DFG) through the Research
Training Group 1462 is gratefully acknowledged. Any opinions, findings, conclusions or
recommendations expressed in this paper are solely those of the authors and do not
necessarily reflect the views of DFG or any other organizations and collaborators.
 S. Jeong, Y. Zhang, J. P. Lynch, H. Soon and K. H. Law. A NoSQL-Based Data
Management Infrastructure for Bridge Monitoring Database. In: Proceedings of the
International Workshop on Structural Health Monitoring. Stanford, CA, USA,
September 1, 2015.
 K. H. Law, K. Smarsly and Y. Wang. Sensor Data Management Technologies for
Infrastructure Asset Management. In: M. L. Wang, J. P. Lynch and H. Sohn (eds.).
Sensor Technologies for Civil Infrastructures, pp. 3-32. Sawston, UK: Woodhead
 D. Hartmann, K. Smarsly and K. H. Law. Coupling Sensor-Based Structural Health
Monitoring with Finite Element Model Updating for Probabilistic Lifetime Estimation
of Wind Energy Converter Structures. In: Proceedings of the International Workshop
on Structural Health Monitoring. Stanford, CA, USA, September 1, 2011.
 K. Smarsly, D. Hartmann and K. H. Law. A Computational Framework for Life-Cycle
Management of Wind Turbines incorporating Structural Health Monitoring. Structural
Health Monitoring An International Journal, 12(4), pp. 359-376, 2013.
 C. R. Farrar and N. A. J. Lieven. Damage prognosis: the future of structural health
monitoring. Philosophical Transactions of the Royal Society A, 365(2007), pp. 623-
 D. An, N. H. Kim and J.-H. Choi. Options for Prognostics Methods: A review of data-
driven and physics-based prognostics. In: Proceedings of the Annual Conference of the
Prognostics and Health Management Society. New Orleans, LA, USA, October 14,
 K. Dragos and K. Smarsly. Decentralized infrastructure health monitoring using
embedded computing in wireless sensor networks. In: A. Sextos and G. D. Manolis
(eds.). Dynamic Response of Infrastructure to Environmentally Induced Loads. In print.
Cham, Switzerland: Springer International Publishing AG, 2016.
 K. Dragos and K. Smarsly. Embedding numerical models into wireless sensor nodes for
structural health monitoring. In: Proceedings of the International Workshop on
Structural Health Monitoring. Stanford, CA, USA, September 1, 2015.
 K. Jahr, R. Schlich, K. Dragos and K. Smarsly. Decentralized autonomous fault
detection in wireless structural health monitoring systems using structural response
data. In: Proceedings of the International Conference on the Applications of Computer
Science and Mathematics in Architecture and Civil Engineering. Weimar, Germany,
July 22, 2015.
 K. Smarsly, K. H. Law and D. Hartmann. A Cyberinfrastructure for Integrated
Monitoring and Life-Cycle Management of Wind Turbines. In: Proceedings of the
International Workshop on Intelligent Computing in Engineering. Vienna, Austria, July
 S. W. Doebling, C. R. Farrar, M. B. Prime and D. W. Shevitz. Damage Identification
and Health Monitoring of Structural and Mechanical Systems from Changes in Their
Vibration Characteristics: A Literature Review. Report. Los Alamos National Lab.,
NM, USA, 1996.
 C. R. Farrar and K. Worden. Structural Health Monitoring A Machine Learning
Perspective. Chichester, UK: John Wiley & Sons, Ltd., 2013.
 K. Worden and G. Manson. The application of machine learning to structural health
monitoring. Philosophical Transactions of the Royal Society A, 365(2007), pp. 515-
 E. Figueiredo, G. Park, C. R. Farrar, K. Worden and J. Figueiras. Machine learning
algorithms for damage detection under operational and environmental variability.
Structural Health Monitoring An International Journal, 10(6), pp. 559-572, 2010.
 N. Dervilis. A machine learning approach to structural health monitoring with a view
towards wind turbines. PhD Thesis. Sheffield, UK: University of Sheffield, 2013.
 J. Park, K. Smarsly, K. H. Law and D. Hartmann. Analyzing the Temporal Variation of
Wind Turbine Responses Using Gaussian Mixture Model and Gaussian Discriminant
Analysis. Journal of Computing in Civil Engineering, 29(2013), 10.1061/(ASCE)CP.
1943-5487.0000416, B4014011, 2013.
 W. Nick, K. Asamene, G. Bullock, A. Esterline and M. Sundaresan. A Study of
Machine Learning Techniques for Detecting and Classifying Structural Damage.
International Journal of Machine Learning and Computing, 5(4), pp. 313-318, 2015.
 I. H. Witten, E. Frank, M. A. Hall. Data mining: Practical machine learning tools and
Edition). Burlington, MA, USA: Morgan Kaufmann Publ., Elsevier
 W. Nick, J. Shelton, K. Asamene and A. Esterline. A study of supervised machine
learning techniques for structural health monitoring. In: Proceedings of 26th Modern AI
and Cognitive Science Conference. Greensboro, NC, USA, April 25, 2015.
 S. Chen, et al. Multiresolution classification with semi-supervised learning for indirect
bridge structure health monitoring. In: Proceedings of the IEEE International
Conference on Acoustics, Speech Signal Processing. Vancouver, Canada, May 26,
 S. B. Kotsiantis. Supervised Machine Learning: A Review of ClassificationTechniques.
Informatica, 31(2007), pp. 249-268, 2007.
 K. Smarsly and Y. Petryna. A Decentralized Approach towards Autonomous Fault
Detection in Wireless Structural Health Monitoring Systems. In: Proceedings of the 7th
European Workshop on Structural Health Monitoring. Nantes, France, July 8, 2014.
 R. Brincker, P. Andersen and L. Zhang. Modal identification from ambient responses
using frequency domain decomposition. In: Proceedings of the 18th International
Modal Analysis Conference (IMAC). San Antonio, TX, USA, February 7, 2000.
 K. Dragos, K. Jahr and K. Smarsly. Nonlinear sensor fault diagnosis in wireless sensor
networks using structural response data. In: Proceedings of the 23rd International
Workshop of the European Group for Intelligent Computing in Engineering. Krakow,
Poland, June 29, 2016.
 K. Smarsly and K. H. Law. Decentralized Fault Detection and Isolation in Wireless
Structural Health Monitoring Systems using Analytical Redundancy. Advances in
Engineering Software, 73(2014), pp. 1-10.